The vvas_xinfer plug-in includes built-in profiling capabilities that allow you to monitor
and analyze the performance of the inference pipeline. To enable profiling, set the enable-profiler
property to true on the vvas_xinfer plugin element.
The profiling data is automatically written to a JSON output file named vvas_xinfer_profiler_data_XXXX.json,
which contains detailed performance metrics for analysis. You can customize the output file name by setting
the profiler-file property to specify a custom file path.
Optionally, set the profiler-log-interval property to define the log frequency in seconds to get periodic
profiling logs at runtime. To view the profiling logs in the console, the inference log level must be set to
GST_INFO or higher.
Example Pipeline with Profiling Enabled:
Note
The following JSON examples are only samples. Ensure
model-pathis set to point to your available model before running.
export GST_DEBUG="vvas_xinfer:4"
gst-launch-1.0 filesrc location=<INPUT>.rgb \
! rawvideoparse width=<INPUT_WIDTH> height=<INPUT_HEIGHT> format=rgb ! queue \
! vvas_xinfer config-file=/etc/vvas/configs/infer/resnet50_vart.json \
enable-profiler=true profiler-log-interval=5 ! queue min-threshold-buffers=4 ! queue \
! vvas_xmetaconvert config-location=/etc/vvas/configs/metaconvert/metaconvert_config.json \
! vvas_xoverlay ! queue ! fpsdisplaysink video-sink="fakesink" text-overlay=false sync=false
In this example, profiling is enabled with a log interval of 5 seconds, allowing periodic performance monitoring during the inference operation.